PixGR: Pixel-Space Generative Decoding for Fast Feed-Forward 3D Reconstruction

Researchers released PixGR, a framework that adds lightweight pixel-space generative modeling to feed-forward 3D Gaussian reconstruction, detailed in arXiv paper 2610.31908v1. PixGR encodes context observations once into a scene latent, then uses pixel-space flow matching with a lightweight DiT decoder to generate a shared 3D Gaussian representation, avoiding VAE-based latent compression and the error propagation of two-stage pipelines. The authors report PixGR outperforms regression-based approaches in sparse-view settings, including widely spaced observations and view extrapolation, while maintaining sub-second inference and remaining competitive with substantially heavier pretrained generative models.

arXiv:2610.31908v1 Announce Type: new Abstract: We present PixGR, an efficient framework that integrates lightweight pixel-space generative modeling into feed-forward 3D Gaussian reconstruction to model scene uncertainty. Context observations are encoded once into a scene latent representation, which conditions generative decoding from noisy target views to produce a shared 3D Gaussian representation. We formulate this process using pixel-space flow matching, enabling end-to-end optimization of 3D reconstruction with direct pixel-space generative supervision through differentiable rendering. This avoids VAE-based latent compression and error propagation inherent in previous two-stage pipelines. To preserve feed-forward efficiency, PixGR caches the scene representation and performs iterative inference using a lightweight DiT decoder. Experiments demonstrate that PixGR outperforms regression-based approaches in challenging sparse-view settings, particularly with widely spaced observations and view extrapolation. Meanwhile, it maintains sub-second inference and remains competitive with substantially heavier pretrained generative models. Our project page is available at https://ranrhuang.github.io/pixgr/.